Segmentation of fMRI Data by Maximization of Region Contrast

Functional segmentation of an fMRI image is the partitioning of the image voxels into clumps that are comodulated by task-related neural activation or other processes. This grouping of voxels can be useful for exploratory analysis and for detection of signal artifacts that can confound regression-based analysis. We present a new method for functional segmentation that is based upon maximization of the contrast of an image-region with respect to its spatial neighborhood. Contrast maximization is an appealing approach since it does not require imagespecific tuning of parameters, as with many clustering approaches. With synthetic benchmark datasets, the accuracy of this Auto-threshold Contrast Enhancing Iterative Clustering (ACEIC) method compares favorably with that of Probabilistic ICA, which is a popular method used for detection of functional groupings of voxels. We also show that functional segmentation of in vivo datasets with ACEIC can provide insights into the results from conventional analysis.